just deep-research-falcon human BIN1 --fallback perplexity-lite timed out after 180 seconds with no generated research artifact, so this manual review uses cached UniProt, GOA, PANTHER family, and publication evidence.protein binding annotations are over-annotated; more informative binding terms such as phospholipid binding, clathrin binding, GTPase binding, and actin filament binding are retained where supported.The second-pass audit added manual reference_review metadata for BIN1 membrane-curvature/BAR-domain evidence, disease-mutant membrane tubulation evidence, DNM2 interaction evidence, BIN1-BACE1 trafficking, and BIN1/tau/clusterin Alzheimer-context interaction evidence. No annotation action changes were needed: BIN1 remains curated primarily as a BAR/SH3 membrane-remodeling and endocytic adaptor, with BACE1/tau/Alzheimer outputs retained as non-core context rather than the defining molecular function.
A Falcon (Edison) deep-research report is now in BIN1-deep-research-falcon.md; it broadly corroborates the existing review's framing of BIN1 as a BAR/SH3 membrane-curvature and endocytic-adaptor protein with non-core muscle T-tubule, cancer/MYC, and Alzheimer roles, adding mechanistic and cardiac detail but no findings that contradict the current annotations. The Falcon-sourced citations below are not yet independently verified against full text.
New or refined findings beyond the existing notes/review:
Discrepancies / annotations to revisit:
Reassessed all 112 existing review objects against the normal notes, UniProt functional/isoform information, all 28 cached abstracts, all 12 Reactome summaries, and selected primary Results/Methods. The 122 raw GOA rows, 112 original annotation source objects and eleven alternative products are unchanged. GOA contains partner sets that the old YAML omitted; the temporary row-to-source projection records those sets without silently backfilling them or splitting existing source objects. Qualifiers, including cytoskeletal colocalizes_with, remain unchanged. No NEW, REMOVE or MARK_AS_OVER_ANNOTATED decision is introduced.
The broad establishment-of-localization assertion is compatible with BIN1; its old over-annotation rationale confused breadth with excessive specificity. Its proposed replacement is the already established endocytosis process. Generic binding assertions were assessed individually. NS5A and REPS2 contacts support proline-rich region binding; MTM1 supports phosphatase binding; DNM2 supports GTPase binding; the CLIP170/nesprin contacts support cytoskeletal partner terms. These terms describe the partner or region recognized by BIN1 and do not give BIN1 the partner's enzyme activity. RIN3 binding is retained as non-core: selected Results leave the precise SH2/proline-rich interface less resolved than the abstract suggests. Mixed proteome-array partners are not forced into a single enzyme class.
The primary membrane study directly tests N-BAR-driven deformation with purified protein and liposomes and N-BAR constructs in C2C12 myoblasts. The membrane-bending core uses GO:0180020 rather than an undifferentiated binding label. Dynamin recruitment and actin/nuclear-envelope coupling form two further cores. The latter paper describes a direct structural mechanism for nuclear positioning; the earlier tumor-suppressor rationale was inapplicable to that annotation. Redundant broad core locations/processes and provider-only supporting quotations were removed. These are evidence-grounded core summaries, with no new standalone GOA assertion.
The BACE1 paper remains abstract-only in the normal cache; publisher article and PDF links returned an abstract and purchase page. Its trafficking and abundance results do not settle the molecular inhibitor annotation, so the earlier confident replacement by several different-aspect terms becomes UNDECIDED. The existing endosome-to-lysosome process is retained as a trafficking function, and its amyloid-beta outcome remains non-core. RNA-polymerase binding is likewise unresolved because the original abstract emphasizes NS5A motif recognition and the exact polymerase/processed-chain interpretation was not independently established. Nuclear fractionation and transcription-repressor-complex membership also retain explicit source-level uncertainty.
Selected human-brain fractionation and binding Results support membrane/cytosolic/cytoskeletal pools and distinguish clusterin-binding splice forms. Selected RIN3 Results distinguish human-cell binding assays from PC12 and mouse-neuron localization. Selected 1997 localization Results concern rat brain/muscle and specified transfected cells, not universal human-tissue localization. The HENA, BioPlex, OpenCell and multimodal exact pair-level data were not recovered; unresolved review decisions do not allege false interactions or missing genes in those datasets. No full-paper, image or supplementary-data audit is claimed.
Official GO pages corroborated binding definitions and the membrane-bending label/MF placement. The full membrane-bending definition and selected ontology relationships were read from the local GO-plus 2026-03-25 snapshot after live definition endpoints failed; this is not an exhaustive current-ontology search. No BIN1/O00499 match was found in the local GO-CAM index. No NEW biological-process term is proposed, so no unperformed comparator check is claimed. The existing generated Falcon report and its nested artifacts remain immutable background, not independent primary evidence. Historical notes are preserved verbatim; new notes paraphrase sources rather than reproduce their earlier long quotations.
This is a temporary scientific proposal awaiting an independent peer. Canonical review/notes, source caches, history and remote state have not been changed by this preparation.
Focused temporary-proposal validation passed with schema, term, reference-text and GOA checks. Six advisories remain: four supported generic interactions retained as non-core, the deliberate source-specific uncertainty for one nuclear-localization experiment, and the membrane-bending core term without a new standalone annotation. The first validation attempt failed to launch validator subprocesses because of environment/cache configuration; the configured offline run passed. This does not claim a repository-wide validation or canonical application.
Independent scientific review accepted this proposal on 2026-10-01 UTC. The canonical review now contains 59 ACCEPT, 34 KEEP_AS_NON_CORE, 10 MODIFY and 9 UNDECIDED decisions across the same 112 source annotations and eleven alternative products. All three core functions were independently checked against the cited primary evidence; source caches and generated provider artifacts remain unchanged. The exact review bytes passed focused schema, term, reference and GOA validation before application; six documented advisories remain.
The current interpretation of the BACE1 study explicitly supersedes the June 19 and June 21 inference that trafficking results settle the aspartic-protease-inhibitor annotation. Trafficking-dependent changes in BACE1 abundance are supported; direct inhibition of BACE1 molecular activity is a separate assertion. The accessible abstract and earlier publisher/PDF inspection do not expose the complete experiment, so they do not demonstrate either the presence or absence of additional inhibition assays. GO:0019828 remains UNDECIDED, while the existing trafficking and amyloid-beta annotations retain their separate scope. The unchanged historical journal and generated Falcon material record the earlier interpretation, not a second current conclusion.
The June statement calling all generic protein-binding assertions over-annotations is also superseded. BIN2, SNX4, the proteome-array partners and RIN3 have positive interaction evidence and remain non-core where a common, defensible finer activity has not been established. The HENA, BioPlex, OpenCell and multimodal records remain UNDECIDED because their exact pair-level evidence was not inspected. The user action definitions distinguish insufficient evidence from a false assertion; generic wording alone does not establish that those experimental interactions are incorrect. This follow-up does not invent a partner-specific mechanism or remove them solely to satisfy a blanket binding-policy recommendation.
The astrocytoma/neuroblastoma study contains two positive, cell-state-dependent results. BIN1 overexpression induces apoptosis in undifferentiated cells; in cells already committed to neuronal or astrocyte differentiation it instead advances differentiated phenotypes. The latter result supports the existing non-core differentiation annotation because GO:0048711 includes an increase in the extent of differentiation, not only its initiation. These are transformed-cell overexpression results, not a claim about universal differentiation or normal human astrocyte development. The two review reasons now state their distinct positive results.
The viral-chain discrepancy is now explicit. The official P27958 feature record maps PRO_0000037576 to NS5A (1973-2420) and PRO_0000037577 to the separate NS5B RNA polymerase (2421-3011). The original BIN1 interaction abstract assays NS5A PxxP recognition. The second GOA partner, Q9WMX2-PRO_0000037551, was not independently resolved through its exact feature record: the text/JSON and IntAct pages were inaccessible and the entry page returned a fallback shell. The prior reason identifying both chains confidently is narrowed accordingly. A whole-paper or polymerase-complex interpretation was not verified, so the RNA-polymerase-binding row remains UNDECIDED; neither a direct NS5B interaction nor a curator error is asserted.
Short exact cache anchors now accompany the DNM2 refinement and the nesprin/CLIP170 refinement. The latter source object combines a cytoskeletal linker and a microtubule regulator, whereas the CLIP170-only study specifically supports the narrower regulatory-partner term. The third core description now uses nucleus organization to encompass the supported positioning and shape mechanism; no new process annotation is added.
This bounded reassessment used the ten cited cached abstracts, selected RIN3 Results text, existing ontology evidence and the new official chain/astrocyte-term checks. It does not claim full-paper or supplementary-pair reinspection. All 112 annotation source objects, their actions and replacement terms, eleven products, references and immutable caches are retained. The earlier temporary-proposal/pending statements are historical and were superseded by the recorded independent application and publication; this appended follow-up is a new temporary proposal awaiting a separate scientific peer.
Independent science review passed for this focused follow-up on 2026-10-01 UTC. The exact approved YAML and appended notes were applied after matching canonical preimages and the published PR head. All 112 source objects, eleven products, actions and replacement terms are preserved. This follow-up clarifies the BACE1 assay-access limit, positive cell-state-dependent differentiation evidence, one confirmed viral chain and one unresolved chain, and binding specificity; it adds short exact evidence anchors and explicitly supersedes historical interpretations. Focused canonical validation, rendering and a new history record follow this application.
Focused canonical validation, HTML rendering and history validation passed on 2026-10-01 UTC. The rendered page embeds the approved YAML. Seven advisories remain: four generic-binding non-core assignments, source-dependent nucleus decisions, no generated-provider citation, and the membrane-bending core term lacking a matching existing-annotation row. Those advisories do not change the bounded evidence decisions or create a new annotation. No full-repository validation is claimed. All normal sources and previous history remain unchanged.
Four positively adjudicated GO:0005515 protein binding rows were migrated from KEEP_AS_NON_CORE to REMOVE to follow the repository policy for generic binding in re-reviewed annotations. BIN2, SNX4, the proteome-array partners and RIN3 remain supported as reported interactions, but GO:0005515 does not record a specific BIN1 function and no narrower molecular-function term is justified by the accessed evidence. Removal therefore reflects term specificity only and does not assert that the interactions are false.
The four high-throughput interactome-screen rows that lack recovered pair-level evidence remain UNDECIDED, matching the evidence-access carve-out in the same policy.
This entry supersedes the four REMOVE decisions from the 2026-10-05 generic-binding migration. The governing authority is the explicit user instruction in the CLINGEN_MENDELIAN project at immutable main revision f7dc8b60: retain a supported, biologically correct GO:0005515 (protein binding) annotation as KEEP_AS_NON_CORE when no evidence-backed, more specific replacement has been established. Evidence-specific uncertainty still requires UNDECIDED; this instruction does not verify an interaction by itself.
Three rows return to KEEP_AS_NON_CORE. The complete BIN2 abstract, PMID:10903846, supports a BAR-dependent cellular BIN2-BIN1 complex; exact BIN1 splice form, purified stoichiometry and a heterodimer-specific mechanism were not reconstructed. For PMID:16275660(https://pubmed.ncbi.nlm.nih.gov/16275660/), the complete abstract plus exact IntAct source records verify all three recorded human partners, FBP1, XRCC4 and DLGAP4, with the isolated BIN1 SH3 region at residues 497-593. Filter-binding, pull-down, two-hybrid and peptide-array records supply Table 1/Figure 4 provenance. This is shared curated provenance, not independent replication. Recorded peptide regions are heterogeneous; inspection of those coordinates in current UniProt sequences does not reconstruct original assay sequence versions. No common proline-rich recognition rule or partner-enzyme activity is inferred, and SH3-mediated recognition is not reversed into SH3-domain binding. The publisher full article and original figure pixels were not successfully retrieved for a new audit; exact pair/assay linkage is the checked scope.
The RIN3 source, PMID:32552912, was read in its complete cached abstract and selected interaction Results, Figure 3 legend, plasmid Methods and co-IP Methods. Those describe HEK293T co-IP and yeast two-hybrid evidence, using a neuronal BIN1 isoform. Its exact UniProt isoform accession remains unestablished. IntAct EBI-26520878/IM-28480-3 and EBI-26529583/IM-28480-7 independently confirm which human pair and assay the curated records describe, while sharing the publication's provenance. PC12 and mouse-neuron recruitment remain separate context. The Results leave the RIN3 SH2/PRD contribution unresolved and did not directly delete the PRD, so no specific BIN1 recognition sequence or RIN3 GEF activity is assigned to BIN1. Figure pixels, full supplements and complete-paper inspection are not claimed.
The SNX4 row instead becomes UNDECIDED. Beyond the complete normal abstract, selected primary PMID:12668730 article text available through ResearchGate was read in the cDNA Methods and Figure 1-2 Results/legends. It identifies mouse Amp2m/Bramp2 and human SNX4 constructs. Endogenous amphiphysin in 3T3 is murine; the Bramp2 co-IP in HeLa likewise uses a mouse construct. Thus HeLa context cannot be represented as a human BIN1 assay. The paper supplies positive cross-species association evidence, but the exact human BIN1 IPI assertion was not resolved. This is not a negative binding result or a claim of curator misattribution. The original evidence code, reference and absent isoform field are preserved. Publisher access was blocked; selected original article text in the mirror was inspected, not original figure pixels or every assay.
Only four review summaries/actions/reasons and this notes appendix change. All 112 source objects, eleven alternative products, three core functions, other annotations, reference assessments and immutable caches remain unchanged. The four previously unresolved survey rows remain UNDECIDED. Current totals are 59 ACCEPT, 33 KEEP_AS_NON_CORE, 10 MODIFY and 10 UNDECIDED; generic-binding totals are nine MODIFY, three KEEP_AS_NON_CORE and five UNDECIDED. Every scientific quotation is unchanged, with aggregate totals at most 22 words per source; no duplicate reduction or new scientific quotation is needed. The original merged PR does not close the campaign hold until this source-specific follow-up passes review/CI and merges.
The current SNX4 assessment is deliberately grounded in the authentic cached abstract of PMID:12668730, DOI:10.1242/jcs.00403: the screen used the N-terminal 1–304 region of murine amphiphysin 2. The abstract reports co-immunoprecipitation confirmation without identifying a human BIN1 construct for that assay. The row therefore remains UNDECIDED for the exact human source assertion. The earlier external Methods/Results reading remains part of this journal, but the named mirror-only constructs and cell-line details are no longer premises of the current annotation rationale. Neither evidence gap is a claim of absent binding or curator error.
The earlier exact-target IntAct query for this paper had returned no records. Fresh publication-only, SNX4-only and exact human-pair queries likewise returned no records. Their query URLs, timestamps and empty-response hashes are recorded in the finite interaction-evidence extract. Empty results cannot determine the experimentally tested organism. The existing GOA assertions and immutable abstract cache are unchanged.
Five positive records were separately retrieved by accession and are reproduced as selected verbatim machine fields in that same JSON, so their PMID, partners, human taxa, method, tags/regions and figure locator are inspectable offline. The original response hashes and resolvable source queries are included. These curated records share the publications' provenance and are not independent experimental replication:
For the array source, GO:0070064 proline-rich region binding was considered explicitly. The accessible source text and recorded heterogeneous partner regions do not establish that refinement for this combined source row. The earlier current-sequence inspection is not an alignment of the original assay constructs. Isolated SH3-domain binding does not establish full-length endogenous complex membership, and no new figure-pixel or complete-paper audit is claimed. Those limits belong here; the annotation reason now states the biological evidence and decision concisely.
All 112 annotation source objects, all actions and replacement terms, eleven alternative products, three core functions, references and cached publications remain unchanged. Only two review summaries/reasons, this appended note and the authentic record extract are added or refined. No new scientific quotations are introduced, and existing supporting quotations are unchanged.